Stanford-led team launches AItonomy Foundation for AI4Science, releases code acceleration benchmarks
seanwbren · x · 2026-08-18
The AItonomy Foundation, founded by PhDs and postdocs from Stanford, Princeton, and Oxford, aims to bridge AI4Science and Science4AI by building open infrastructure. They launched two initial projects:
- ScienceAccelBench: A benchmark to evaluate AI's ability to port trusted scientific code (e.g., Fortran/C solvers) to GPUs without altering results. Scientists contribute code and correctness criteria; AI systems produce ports, which are verified against original outputs. The benchmark sets a high bar, requiring current agents like Claude Code and Codex to fail on specific tasks.
- ScienceMysteryBench: Rewinds finished discoveries to raw data to test AI's ability to replicate scientific breakthroughs.
The organization commits to open-sourcing all outputs to accelerate progress in both science and AI.
More from coding & agent
- Grokbot Tutorial: Building 5 Practical AI Agents for Personal Use — petergyang · 2026-08-18
- Paper: A Programming Paradigm for Spatiotemporal Composability — sull · 2026-08-18
- The State of Vibe Coding: Cracks Hard Problems Like a Charm, Then Deletes Your Production Database — 2C_ornot2C · 2026-08-18
- Understanding Next-Gen AI Agents: From Memory to Self-Evolving Systems — sven_ai · 2026-08-18
- blind_watermark: 14k-star open-source tool embeds invisible watermarks that survive resize, crop and blur — lxfater · 2026-08-18
- Dev Updates Website UI to Match iOS App in Under an Hour with Replit — amasad · 2026-08-18